Make index and global metadata upload timeout dynamic cluster settings (

#10814) (#10852)

* Make index and global metadata upload wait time dynamic

Signed-off-by: Rahul Karajgikar <[email protected]>
(cherry picked from commit 8f13dee)

CHANGELOG.md

@@ -12,6 +12,7 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 - [Remote cluster state] Upload global metadata in cluster state to remote store([#10404](https://github.com/opensearch-project/OpenSearch/pull/10404))
 - [Remote cluster state] Download functionality of global metadata from remote store ([#10535](https://github.com/opensearch-project/OpenSearch/pull/10535))
 - [Remote cluster state] Restore global metadata from remote store when local state is lost after quorum loss ([#10404](https://github.com/opensearch-project/OpenSearch/pull/10404))
+- [Remote cluster state] Make index and global metadata upload timeout dynamic cluster settings ([#10814](https://github.com/opensearch-project/OpenSearch/pull/10814))
 - Add search query categorizor ([#10255](https://github.com/opensearch-project/OpenSearch/pull/10255))
 
 ### Dependencies

server/src/main/java/org/opensearch/common/settings/ClusterSettings.java

@@ -681,6 +681,8 @@ public void apply(Settings value, Settings current, Settings previous) {
 
                 // Remote cluster state settings
                 RemoteClusterStateService.REMOTE_CLUSTER_STATE_ENABLED_SETTING,
+                RemoteClusterStateService.INDEX_METADATA_UPLOAD_TIMEOUT_SETTING,
+                RemoteClusterStateService.GLOBAL_METADATA_UPLOAD_TIMEOUT_SETTING,
                 RemoteStoreNodeService.REMOTE_STORE_COMPATIBILITY_MODE_SETTING,
                 IndicesService.CLUSTER_REMOTE_TRANSLOG_BUFFER_INTERVAL_SETTING,
                 IndicesService.CLUSTER_REMOTE_INDEX_RESTRICT_ASYNC_DURABILITY_SETTING

server/src/main/java/org/opensearch/gateway/remote/RemoteClusterStateService.java

@@ -83,9 +83,23 @@ public class RemoteClusterStateService implements Closeable {
 
     private static final Logger logger = LogManager.getLogger(RemoteClusterStateService.class);
 
-    // TODO make this two variable as dynamic setting [issue: #10688]
-    public static final int INDEX_METADATA_UPLOAD_WAIT_MILLIS = 20000;
-    public static final int GLOBAL_METADATA_UPLOAD_WAIT_MILLIS = 20000;
+    public static final TimeValue INDEX_METADATA_UPLOAD_TIMEOUT_DEFAULT = TimeValue.timeValueMillis(20000);
+
+    public static final TimeValue GLOBAL_METADATA_UPLOAD_TIMEOUT_DEFAULT = TimeValue.timeValueMillis(20000);
+
+    public static final Setting<TimeValue> INDEX_METADATA_UPLOAD_TIMEOUT_SETTING = Setting.timeSetting(
+        "cluster.remote_store.state.index_metadata.upload_timeout",
+        INDEX_METADATA_UPLOAD_TIMEOUT_DEFAULT,
+        Setting.Property.Dynamic,
+        Setting.Property.NodeScope
+    );
+
+    public static final Setting<TimeValue> GLOBAL_METADATA_UPLOAD_TIMEOUT_SETTING = Setting.timeSetting(
+        "cluster.remote_store.state.global_metadata.upload_timeout",
+        GLOBAL_METADATA_UPLOAD_TIMEOUT_DEFAULT,
+        Setting.Property.Dynamic,
+        Setting.Property.NodeScope
+    );
 
     public static final ChecksumBlobStoreFormat<IndexMetadata> INDEX_METADATA_FORMAT = new ChecksumBlobStoreFormat<>(
         "index-metadata",
@@ -141,6 +155,9 @@ public class RemoteClusterStateService implements Closeable {
     private BlobStoreTransferService blobStoreTransferService;
     private volatile TimeValue slowWriteLoggingThreshold;
 
+    private volatile TimeValue indexMetadataUploadTimeout;
+    private volatile TimeValue globalMetadataUploadTimeout;
+
     private final AtomicBoolean deleteStaleMetadataRunning = new AtomicBoolean(false);
 
     public static final int INDEX_METADATA_CURRENT_CODEC_VERSION = 1;
@@ -171,7 +188,11 @@ public RemoteClusterStateService(
         this.relativeTimeNanosSupplier = relativeTimeNanosSupplier;
         this.threadpool = threadPool;
         this.slowWriteLoggingThreshold = clusterSettings.get(SLOW_WRITE_LOGGING_THRESHOLD);
+        this.indexMetadataUploadTimeout = clusterSettings.get(INDEX_METADATA_UPLOAD_TIMEOUT_SETTING);
+        this.globalMetadataUploadTimeout = clusterSettings.get(GLOBAL_METADATA_UPLOAD_TIMEOUT_SETTING);
         clusterSettings.addSettingsUpdateConsumer(SLOW_WRITE_LOGGING_THRESHOLD, this::setSlowWriteLoggingThreshold);
+        clusterSettings.addSettingsUpdateConsumer(INDEX_METADATA_UPLOAD_TIMEOUT_SETTING, this::setIndexMetadataUploadTimeout);
+        clusterSettings.addSettingsUpdateConsumer(GLOBAL_METADATA_UPLOAD_TIMEOUT_SETTING, this::setGlobalMetadataUploadTimeout);
     }
 
     private BlobStoreTransferService getBlobStoreTransferService() {
@@ -367,7 +388,7 @@ private String writeGlobalMetadata(ClusterState clusterState) throws IOException
         );
 
         try {
-            if (latch.await(GLOBAL_METADATA_UPLOAD_WAIT_MILLIS, TimeUnit.MILLISECONDS) == false) {
+            if (latch.await(getGlobalMetadataUploadTimeout().millis(), TimeUnit.MILLISECONDS) == false) {
                 // TODO: We should add metrics where transfer is timing out. [Issue: #10687]
                 GlobalMetadataTransferException ex = new GlobalMetadataTransferException(
                     String.format(Locale.ROOT, "Timed out waiting for transfer of global metadata to complete")
@@ -422,7 +443,7 @@ private List<UploadedIndexMetadata> writeIndexMetadataParallel(ClusterState clus
         }
 
         try {
-            if (latch.await(INDEX_METADATA_UPLOAD_WAIT_MILLIS, TimeUnit.MILLISECONDS) == false) {
+            if (latch.await(getIndexMetadataUploadTimeout().millis(), TimeUnit.MILLISECONDS) == false) {
                 IndexMetadataTransferException ex = new IndexMetadataTransferException(
                     String.format(
                         Locale.ROOT,
@@ -621,6 +642,22 @@ private void setSlowWriteLoggingThreshold(TimeValue slowWriteLoggingThreshold) {
         this.slowWriteLoggingThreshold = slowWriteLoggingThreshold;
     }
 
+    private void setIndexMetadataUploadTimeout(TimeValue newIndexMetadataUploadTimeout) {
+        this.indexMetadataUploadTimeout = newIndexMetadataUploadTimeout;
+    }
+
+    private void setGlobalMetadataUploadTimeout(TimeValue newGlobalMetadataUploadTimeout) {
+        this.globalMetadataUploadTimeout = newGlobalMetadataUploadTimeout;
+    }
+
+    public TimeValue getIndexMetadataUploadTimeout() {
+        return this.indexMetadataUploadTimeout;
+    }
+
+    public TimeValue getGlobalMetadataUploadTimeout() {
+        return this.globalMetadataUploadTimeout;
+    }
+
     static String getManifestFileName(long term, long version, boolean committed) {
         // 123456789012_test-cluster/cluster-state/dsgYj10Nkso7/manifest/manifest__<inverted_term>__<inverted_version>__C/P__<inverted__timestamp>__<codec_version>
         return String.join(

server/src/test/java/org/opensearch/gateway/remote/RemoteClusterStateServiceTests.java

@@ -102,6 +102,7 @@
 public class RemoteClusterStateServiceTests extends OpenSearchTestCase {
 
     private RemoteClusterStateService remoteClusterStateService;
+    private ClusterSettings clusterSettings;
     private Supplier<RepositoriesService> repositoriesServiceSupplier;
     private RepositoriesService repositoriesService;
     private BlobStoreRepository blobStoreRepository;
@@ -132,6 +133,7 @@ public void setup() {
             .put(RemoteClusterStateService.REMOTE_CLUSTER_STATE_ENABLED_SETTING.getKey(), true)
             .build();
 
+        clusterSettings = new ClusterSettings(settings, ClusterSettings.BUILT_IN_CLUSTER_SETTINGS);
         NamedXContentRegistry xContentRegistry = new NamedXContentRegistry(
             Stream.of(
                 NetworkModule.getNamedXContents().stream(),
@@ -149,7 +151,7 @@ public void setup() {
             "test-node-id",
             repositoriesServiceSupplier,
             settings,
-            new ClusterSettings(settings, ClusterSettings.BUILT_IN_CLUSTER_SETTINGS),
+            clusterSettings,
             () -> 0L,
             threadPool
         );
@@ -1053,6 +1055,38 @@ public void testSingleConcurrentExecutionOfStaleManifestCleanup() throws Excepti
         assertBusy(() -> assertEquals(1, callCount.get()));
     }
 
+    public void testIndexMetadataUploadWaitTimeSetting() {
+        // verify default value
+        assertEquals(
+            RemoteClusterStateService.INDEX_METADATA_UPLOAD_TIMEOUT_DEFAULT,
+            remoteClusterStateService.getIndexMetadataUploadTimeout()
+        );
+
+        // verify update index metadata upload timeout
+        int indexMetadataUploadTimeout = randomIntBetween(1, 10);
+        Settings newSettings = Settings.builder()
+            .put("cluster.remote_store.state.index_metadata.upload_timeout", indexMetadataUploadTimeout + "s")
+            .build();
+        clusterSettings.applySettings(newSettings);
+        assertEquals(indexMetadataUploadTimeout, remoteClusterStateService.getIndexMetadataUploadTimeout().seconds());
+    }
+
+    public void testGlobalMetadataUploadWaitTimeSetting() {
+        // verify default value
+        assertEquals(
+            RemoteClusterStateService.GLOBAL_METADATA_UPLOAD_TIMEOUT_DEFAULT,
+            remoteClusterStateService.getGlobalMetadataUploadTimeout()
+        );
+
+        // verify update global metadata upload timeout
+        int globalMetadataUploadTimeout = randomIntBetween(1, 10);
+        Settings newSettings = Settings.builder()
+            .put("cluster.remote_store.state.global_metadata.upload_timeout", globalMetadataUploadTimeout + "s")
+            .build();
+        clusterSettings.applySettings(newSettings);
+        assertEquals(globalMetadataUploadTimeout, remoteClusterStateService.getGlobalMetadataUploadTimeout().seconds());
+    }
+
     private void mockObjectsForGettingPreviousClusterUUID(Map<String, String> clusterUUIDsPointers) throws IOException {
         mockObjectsForGettingPreviousClusterUUID(clusterUUIDsPointers, false);
     }